KR20150115283A - System and Method of traffic forecasting - Google Patents

Abstract

The present invention relates to a traffic forecasting system and a method thereof. The traffic forecasting system of the present invention includes: an adjacent road collection part which is provided with a target road, and then selects adjacent roads according to adjacency with the target road; a filter part which filters an attentive road among the collected adjacent roads; and a forecasting part which generates traffic information at a forecasted time of the target road by a mechanical learning algorithm, through being provided with traffic information at the past time of the attentive adjacent road and traffic information at the past time of the target road.

Description

[0001] The present invention relates to a traffic prediction system,

The present invention relates to a traffic prediction system and method.

Big data refers to a vast amount of data that is difficult to collect, store, search, and analyze in the conventional way, because the amount, period, and format of the data are too large for existing data. Big data appeared due to the development of various sensors and internet. As computers and processing technologies develop, it becomes possible to observe and predict phenomena based on the big data generated in the digital environment.

Traffic forecasting can also use these big data. Analysis of traffic information using vast past traffic history data can be used to estimate the traffic congestion at the future time point and to be used when judging the optimum route of travel. Therefore, it is important to develop a system and method for accurately analyzing traffic information.

Korean Patent Publication No. 10-2006-0037481

A problem to be solved by the present invention is to provide a more accurate traffic prediction system by simply adding traffic information on an object road as well as information on an adjacent road.

Another problem to be solved by the present invention is to provide a more accurate traffic prediction method by simply adding traffic information on an object road as well as information on an adjacent road.

The problems to be solved by the present invention are not limited to the above-mentioned problems, and other matters not mentioned can be clearly understood by those skilled in the art from the following description.

According to an aspect of the present invention, there is provided a traffic prediction system comprising: an adjacent road collection unit for receiving an object road and selecting an adjacent road according to whether the object road is adjacent to the object road; A filter unit that selects adjacent roads, traffic information of past time points of the neighboring roads, and traffic information of past time points of the target roads, and generates traffic information at a prediction time point of the target road by a machine learning algorithm .

The proximity may be determined by at least one of a link depth to the target road, a distance to the target road, and a travel time to reach the target road.

The significant adjacent roads can be selected by a wrapper method.

The contiguous roads may be selected by a support vector machine (SVM) scheme.

The significant adjacent roads may be selected by a filter method.

The neighboring roads may be selected by a correlation-based feature selection (CFS) method.

The past traffic information of the neighboring roads, the past traffic information of the current road, and the traffic information of the target road may include average speed information.

The machine learning algorithm may be any one of a k-nearest neighbor (k-NN) algorithm, a linear regression or an M5 model tree.

The information processing apparatus may further include a traffic information providing unit for collecting traffic information in real time and providing the real-time traffic information to the adjacent road collection unit and the filter unit.

According to another aspect of the present invention, there is provided a traffic prediction method comprising: receiving an object road and a prediction time point, collecting adjacent roads adjacent to the object road, (K-nearest neighbor) algorithm. The k-nearest neighbor (KNN) algorithm is used to select the neighboring roads according to the k- Lt; / RTI >

The collecting of the adjacent roads may include determining whether the adjacent roads are adjacent according to a link depth with the target road, a distance with the target road, a travel time to reach the target road, or a combination thereof Lt; / RTI >

Selection of the significant adjacent roads may include selection by a support vector machine (SVM) scheme.

The details of other embodiments are included in the detailed description and drawings.

The embodiments of the present invention have at least the following effects.

That is, the traffic prediction system and method according to an embodiment of the present invention can estimate the traffic situation more precisely because the information on the adjacent road as well as the target road is also considered.

In addition, the traffic prediction system and method according to an exemplary embodiment of the present invention can provide a traffic prediction system and method with high precision excluding adjacent roads with low relevance.

The effects according to the present invention are not limited by the contents exemplified above, and more various effects are included in the specification.

1 is a block diagram illustrating a structure of a traffic prediction system according to an embodiment of the present invention.
FIG. 2 is an exemplary diagram for explaining the selection of the adjacent roads based on the depth / spread of the neighbor road collection unit of the traffic prediction system according to the embodiment of the present invention.
FIG. 3 is an exemplary view for explaining the selection of an adjacent road based on the proximity road collection distance of the traffic prediction system according to an exemplary embodiment of the present invention.
FIG. 4 is a diagram illustrating an example of selecting an adjacent road based on a travel time of a neighbor road collection unit of a traffic prediction system according to an exemplary embodiment of the present invention.
FIG. 5 is an exemplary diagram illustrating a method for selecting a neighboring road by a filter unit of a traffic prediction system according to an embodiment of the present invention.
FIG. 6 is an exemplary diagram illustrating a support vector machine (SVM) scheme used by a filter unit of a traffic prediction system according to an exemplary embodiment of the present invention to select a neighboring road.
FIGS. 7 and 8 are conceptual diagrams for explaining a method of predicting traffic information through a prediction-part machine learning algorithm of a traffic prediction system according to an embodiment of the present invention.
9 is a flowchart illustrating a traffic prediction method according to an embodiment of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS The advantages and features of the present invention and the manner of achieving them will become apparent with reference to the embodiments described in detail below with reference to the accompanying drawings. The present invention may, however, be embodied in many different forms and should not be construed as being limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art. Is provided to fully convey the scope of the invention to those skilled in the art, and the invention is only defined by the scope of the claims. Like reference numerals refer to like elements throughout the specification.

Although the first, second, etc. are used to describe various elements, components and / or sections, it is needless to say that these elements, components and / or sections are not limited by these terms. These terms are only used to distinguish one element, element or section from another element, element or section. Therefore, it goes without saying that the first element, the first element or the first section mentioned below may be the second element, the second element or the second section within the technical spirit of the present invention.

It is to be understood that when an element or layer is referred to as being "on" or " on "of another element or layer, All included. On the other hand, a device being referred to as "directly on" or "directly above " indicates that no other device or layer is interposed in between.

The terms spatially relative, "below", "beneath", "lower", "above", "upper" May be used to readily describe a device or a relationship of components to other devices or components. Spatially relative terms should be understood to include, in addition to the orientation shown in the drawings, terms that include different orientations of the device during use or operation. For example, when inverting an element shown in the figure, an element described as " below or beneath "of another element may be placed" above "another element. Thus, the exemplary term "below" can include both downward and upward directions. The elements can also be oriented in different directions, in which case spatially relative terms can be interpreted according to orientation.

The terminology used herein is for the purpose of illustrating embodiments and is not intended to be limiting of the present invention. In the present specification, the singular form includes plural forms unless otherwise specified in the specification. It is noted that the terms "comprises" and / or "comprising" used in the specification are intended to be inclusive in a manner similar to the components, steps, operations, and / Or additions.

Unless defined otherwise, all terms (including technical and scientific terms) used herein may be used in a sense commonly understood by one of ordinary skill in the art to which this invention belongs. Also, commonly used predefined terms are not ideally or excessively interpreted unless explicitly defined otherwise.

Referring to FIG. 1, a traffic prediction system according to an embodiment of the present invention will be described.

1 is a block diagram illustrating a structure of a traffic prediction system according to an embodiment of the present invention.

Referring to FIG. 1, a traffic prediction system according to an embodiment of the present invention includes a traffic information providing unit 210, an adjacent road collection unit 220, a filter unit 230, and a prediction unit 240.

The traffic information providing unit 210 may collect real-time traffic information. The traffic information providing unit 210 may collect real-time traffic information from various sources. The source may include, for example, at least one of a dedicated short range communication (DSRC), a taxi, and a bus.

The traffic information providing unit 210 may transmit the real time traffic information to the adjacent road collection unit 220 and the filter unit 230. [ The real-time traffic information may be updated at a predetermined time interval. For example, the time interval may be 5 minutes, 1 hour, or 4 hours.

Traffic information can be various information. That is, it can be represented by various types of information such as an average speed, a maximum speed, a minimum speed and a running time. The traffic information of the traffic prediction system according to an exemplary embodiment of the present invention may be an average speed.

The adjacent road collection unit 220 may receive the target road from the user 100. [ The target road is a road on which the user 100 wants traffic information. The adjacent road collection unit 220 may collect adjacent roads of the target road provided by the user 100. [ The adjacent road collecting unit 220 considers not only the target road itself but also the traffic information of adjacent roads, so that more accurate traffic prediction can be performed.

The neighbor road collecting unit 220 may receive the real time traffic information from the traffic information providing unit 210. [ The adjacent road collection unit 220 can select adjacent roads according to real-time traffic information.

The filter unit 230 may select the adjacent adjacent roads among the adjacent roads selected from the adjacent road collecting unit 220. The filter unit 230 can receive information on the adjacent road from the adjacent road collecting unit 220. [ The filter unit 230 may also receive real-time traffic information from the traffic information provider 210. The filter unit 230 may determine the relevance of the adjacent road based on the real-time traffic information, and exclude adjacent roads having low relevance. The filter unit 230 can select adjacent roads other than the adjacent roads that are excluded as the adjacent adjacent roads.

The prediction unit 240 may receive the target road from the user 100. [ The prediction unit 240 may also be provided with a prediction time point from the user 100. The prediction time at this time is a time point when the user 100 desires traffic information. That is, the predictor 240 can predict traffic information of a desired road at a desired time point by the user 100. [

The prediction unit 240 may be provided with a significant adjacent road from the filter unit 230. [ The prediction unit 240 can predict the traffic information at the prediction time point of the target road using the traffic information of the past time point of the adjacent road. The predicting unit 240 can also use the past traffic information of the target road to predict the traffic information at the predicted time of the target road. The past traffic information of the adjacent adjacent roads and the past traffic information of the current road may include an average speed.

The prediction unit 240 may provide the user 100 with traffic information at the prediction time of the target road. The traffic information at the predicted time of the target road may include an average speed.

2 to 4, the adjacent road collection unit 220 of the traffic prediction system according to an embodiment of the present invention will be described in detail.

FIG. 2 is a diagram illustrating an example of selecting an adjacent road based on a depth / spread of a neighbor road collection unit of a traffic prediction system according to an exemplary embodiment of the present invention. FIG. FIG. 7 is an exemplary diagram for explaining the selection of the adjacent roads based on the proximity road collection distance of the traffic prediction system according to the embodiment. FIG. FIG. 4 is a diagram illustrating an example of selecting an adjacent road based on a travel time of a neighbor road collection unit of a traffic prediction system according to an exemplary embodiment of the present invention.

The adjacent road collection unit 220 can select the adjacent roads of the target road. The method of selecting the adjacent roads of the target roads is not particularly limited. The adjacent road collection unit 220 may use a method of selecting an adjacent road based on a depth / spread, a method of selecting an adjacent road based on a distance, or a method of selecting an adjacent road based on a driving time have.

Referring to FIG. 2, the adjacent road collecting unit 220 can select an adjacent road based on the depth / spread. The adjacent road collecting unit 220 may preferentially select all the roads that are directly linked to the red target road as adjacent roads (black in [1] in FIG. 2). Next, all the roads connected directly to the previously selected adjacent roads can be selected as adjacent roads ([2] in FIG. 2). Next, all the roads connected directly to the selected adjacent roads can be selected as adjacent roads ([3] in FIG. 2). If the depth is 1, it stops at [1] of FIG. 2. If the depth is 3, the adjacent road can be selected up to [3] of FIG. Therefore, adjacent roads can be selected within a preset depth range.

Referring to FIG. 3, the adjacent road collection unit 220 may select adjacent roads based on the distance. The adjacent road collection unit 220 can select all the roads existing at a predetermined distance from the red target road as adjacent roads. The criteria for measuring the distance may be based on the midpoint of the road. The middle point of the road is the point of the starting point and the end point of the road. That is, the middle point of each road can be determined, and the adjacent road can be selected based on the distance between the midpoint of the target road and the midpoint between the respective roads.

The predetermined distance may be, for example, 300 m, 500 m, or 1000 m. In Fig. 3 [1], the predetermined distance is 300 m. In Fig. 3 [2], the predetermined distance is 500 m, and in [3] of Fig. 3, the predetermined distance is 1000 m.

Referring to FIG. 4, the adjacent road collecting unit 220 can select the adjacent road based on the traveling time. The adjacent road collecting unit 220 defines the time taken to arrive at each road starting from the red target road or the time taken to arrive from each road to arrive at the red target road as the driving time . The adjacent road collecting unit 220 can select a road that is within the predetermined time as the traveling time as an adjacent road.

The predetermined time may be, for example, one minute, two minutes, or three minutes. In Fig. 4 [1], the predetermined time is one minute. In Fig. 4 [2], the predetermined time is 2 minutes. In [3] of FIG. 4, the preset time is three minutes.

Referring to FIG. 5, the filter unit 230 of the traffic prediction system according to an embodiment of the present invention will be described in detail.

FIG. 5 is an exemplary diagram illustrating a method for selecting a neighboring road by a filter unit of a traffic prediction system according to an embodiment of the present invention.

5, the filter unit 230 may select a significant adjacent road as a candidate for the adjacent roads B, C, and D selected by the object road A and the adjacent road collecting unit 220 (I) . Each of the adjacent roads (B, C, D) has an association with the object road, and thus the association with the object road can be calculated. As shown exemplarily in Fig. 5, each adjacent road B, C, D can display a numerical relation with the object road A (II).

Then, the adjacent road C with a low correlation can be removed. At this time, it is possible to sequentially arrange adjacent high-relevancy roads and to select a predetermined number of adjacent roads sequentially from the highest-related neighboring roads. That is, the number of adjacent roads excluded is not particularly limited.

The filter unit 230 may use a wrapper method or a filter method when calculating the relevance and selecting the adjacent adjacent roads.

The wrapper method is a method of performing a traffic prediction algorithm on each of an object road and an adjacent road, respectively, and comparing the resultant values according to the respective algorithms to select the nearest neighbor road. According to such a wrapper method, a precise and error-free result can be obtained.

The filter method is a method of removing at least the adjacent roads at the most disadvantage by comparing at least one parameter that can express the association without performing the traffic prediction algorithm for each of the object road and the adjacent road. According to this filter method, the amount of computation is small, and the result can be obtained quickly.

Referring to FIG. 6, a description will be given of a method in which the filter unit selects the adjacent roads according to the SVM scheme, which is one of the wrapper schemes.

FIG. 6 is an exemplary diagram illustrating a support vector machine (SVM) scheme used by a filter unit of a traffic prediction system according to an exemplary embodiment of the present invention to select a neighboring road.

Referring to FIG. 6, it is possible to have various linear separators for dividing black and white (FIG. 6 (a)). 6 (b)). In other words, the filter unit 230 determines the distribution of the results by the machine learning algorithm among the plurality of adjacent roads A method of determining a significant adjacent road by dividing the largest margin can be used. When the SVM scheme is used, the filter unit 230 can receive the prediction value from the prediction unit 240. [

Referring again to FIG. 5, the filter unit 230 may select a neighboring road by a filter method. The filter method does not calculate the number of all cases like the above-mentioned wrapper method but sequentially selects the significant adjacent roads. Such a filter scheme may include any of forward selection, backward selection, genetic algorithm, and scatter search.

The filter unit 230 may use a correlation-based feature selection (CFS) scheme, which is one of forward selection. However, the present invention is not limited thereto. The CFS scheme may be a scheme of maximizing the correlation and minimizing the redundancy.

The CFS method calculates the correlation coefficient between each adjacent road and the target road.

[Equation 1]

Here, X is a value of an average speed along each time of the object road, and Y is a value of an average speed along each time of the adjacent road. E [X] and E [Y] are expected values of X and Y, respectively. The value of Equation (1) is the covariance of X and Y.

If the covariance is positive, it means that X and Y both increase or decrease together, and if the covariance is negative, X and Y move in opposite directions.

 &Quot; (2) "

This covariance is divided by the standard deviation of X and Y, which is a value obtained by dividing the covariance by the equation (2), which is called a correlation coefficient and has a value between -1 and 1.

The filter unit 230 can search for adjacent roads having the maximum correlation and select the adjacent adjacent roads.

Referring to FIGS. 7 and 8, the predictor of the traffic prediction system according to an embodiment of the present invention will be described in detail.

FIGS. 7 and 8 are conceptual diagrams for explaining a method of predicting traffic information through a prediction-part machine learning algorithm of a traffic prediction system according to an embodiment of the present invention.

Referring to FIGS. 7 and 8, the prediction unit 240 of the present invention can use a k-NN scheme among the machine learning algorithms. However, the present invention is not limited thereto, and a linear regression or an M5 model tree can be used. The k-NN algorithm is described below.

The k-NN algorithm is an algorithm for selecting k data sets having a high correlation with a target data set in a plurality of data sets. In the traffic prediction system according to an embodiment of the present invention, the traffic information of the past time point of the target road is compared with the traffic information of the past time point of the neighboring road, and the most adjacent neighbor roads having the highest correlation are selected.

Referring to FIG. 7, if the current time is t, the past time point can be defined from tn to t-1, and the future time point can be defined from t + 1 to t + m. At this time, the prediction time of the target road desired by the user 100 may be any one of the future time points t + 1 to t + m. For this purpose, past traffic information of the adjacent roads, for example, the average speed, may be used. The average speed of the point of time t-1 of the i-th adjacent point among the various adjacent roads can be expressed as X _{i , t-1} . Here, X represents traffic information such as an average speed, i is a variable for identifying an identification number of a road, that is, a significant adjacent road or an object road, and t-1 represents a time point as described above.

8, an association distance between the data sets 310, 320, 330, and 340 of the adjacent adjacent roads selected by the adjacent road collection unit 220 and the filter unit 230 and the data set 300 of the target road . Here, 'relevance distance' is a measure for determining how much the two objects are related to each other, and is defined as Equation (3). The data sets 310, 320, 330, 340 of the adjacent proximity roads may be data sets of different significance proximity roads, or may be data sets of the same point of time on different dates of the same proximity roads.

&Quot; (3) "

Equation 3 is a formula for finding the association distance between the data set 300 of the object road and the data sets 310, 320, 330, and 340 of the respective adjacent roads. Where _Xj is the object road and _Xq is the significant adjacent road. X _{j , i} is the traffic information of the target road at the _i- th viewpoint, that is, the average speed, and X _{q , i} is the traffic information of the neighboring road at the _i- th point, that is, the average speed. At this time, as the q value changes, the data sets of the adjacent adjacent roads may change with each other.

&Quot; (4) "

Equation 4 is a weight normalized to the association distance between the data set 300 of the target road and the data sets 310, 320, 330, and 340 of the respective adjacent roads. d _j is the association distance between the data set 310, 320, 330, 340 of the nearest neighbor road and the data set 300 of the target road at the jth nearest distance from the data set 300 of the target road. k is a parameter for determining a data set of several adjacent roads as a preset parameter. That is, when k = 1, only one data set 320 of the neighboring roads with the smallest association distance is selected, and the weight is 1. When k is 1 or more, w _j can be obtained by using the distance d _j and the k-th association distance d _k .

&Quot; (5) "

Since w _i is normalized as in Equation (5), the sum of all w _i is 1.

&Quot; (6) "

Equation (6) is a value indicating an average speed prediction value. and y _i is the average speed of the data set of the neighboring roads of the i-th nearest neighboring distance.

That is, when k = 3, only the data sets 310, 320, and 330 of the three neighboring roads having the closest association distance to the data set 300 of the target road are used, The data set 340 can be excluded and the traffic information of the target road at the prediction time point, that is, the average speed, can be obtained.

The data sets 310, 320, and 330 of each of the three adjacent proximity roads have average speeds of 75, 70, and 68, respectively, with respective associated distances of 0.4, 0.25, and 0.35. Therefore, in the case of FIG. 8, the average speed of the predicted time of the target road is 68 * 0.35 + 70 * 0.25 + 75 * 0.4 = 71.3.

That is, the predicting unit 240 can estimate the traffic information at the predicted time of the target road, that is, the average speed, through the process described above.

The k-NN algorithm described above is merely an example of a machine learning algorithm of a traffic prediction system according to an embodiment of the present invention. It is preferable to predict by the M5 model tree method when the adjacent adjacent roads are selected by the filter method, particularly, the CFS method. However, the present invention is not limited thereto. It is preferable to predict by the k-NN method when the significant adjacent roads are selected by the wrapper method, in particular, the SVM method. However, the present invention is not limited thereto.

The traffic prediction system according to an embodiment of the present invention can predict more precise traffic information by using information on adjacent roads. Furthermore, only the significant adjacent roads among these adjacent roads can be selected to further increase the precision.

1 to 9, a traffic prediction method according to an embodiment of the present invention will be described. The portions overlapping with the above description will be simplified or omitted.

1 and 9 are flowcharts for explaining a traffic prediction method according to an embodiment of the present invention.

Referring to FIGS. 1 and 9, the traffic prediction system receives the target road and the prediction time from the user (S100).

The prediction time point may be a future point in time since the user 100 desires prediction.

Next, referring to FIGS. 1 to 4 and 9, an adjacent road collection unit collects adjacent roads (S200).

There are many ways to collect adjacent roads. For example, adjacent roads can be collected by a method according to the depth connected with the target road, a method according to the distance to the target road, and a method according to the running time with the target road.

Next, referring to FIGS. 1, 5, 6, and 9, the filter unit selects the adjacent adjacent roads (S300).

The method of selecting the adjacent roads may be a wrapper method or a filter method. The wrapping method can use the SVM method, and the filter method can use the CFS method of the forward selection method.

1 and 7 to 9, the prediction unit generates traffic information on the target road at the prediction time (S400).

While the present invention has been described in connection with what is presently considered to be practical exemplary embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, You will understand. It is therefore to be understood that the above-described embodiments are illustrative in all aspects and not restrictive.

100: user 200: traffic prediction system
210: traffic information providing unit 220: adjacent road collecting unit
230: filter unit 240: prediction unit

Claims (12)

An adjacent road collection unit for receiving an object road and selecting an adjacent road according to whether the object road is adjacent to the object road;
A filter unit for filtering the neighboring roads among the collected adjacent roads; And
And a predictor configured to receive traffic information of past time points of the neighboring roads and traffic information of past time points of the target roads and to generate traffic information at a prediction time point of the target road by a machine learning algorithm. The method according to claim 1,
Wherein the proximity is determined by at least one of a link depth to the target road, a distance to the target road, and a travel time to reach the target road. The method according to claim 1,
Wherein the significant adjacent roads are selected by a wrapper method. The method of claim 3,
Wherein the significant adjacent roads are selected by a support vector machine (SVM) scheme. The method according to claim 1,
Wherein the significant adjacent roads are selected by a filter method. 6. The method of claim 5,
Wherein the neighboring roads are selected by a correlation-based feature selection (CFS) method. The method according to claim 1,
Wherein traffic information at a past time point of the neighboring road, traffic information at a past time point of the target road, and traffic information at a prediction time point of the target road include average speed information. The method according to claim 1,
Wherein the machine learning algorithm is one of a k-nearest neighbor (k-NN) algorithm, a linear regression or an M5 model tree. The method according to claim 1,
And a traffic information providing unit for collecting traffic information in real time and providing the real-time traffic information to the adjacent road collection unit and the filter unit. The target road and the forecasting point,
Collecting adjacent roads adjacent to the target road,
A neighboring road is selected according to the association with the target road among the adjacent roads,
(K-NN) algorithm to generate the target road traffic information at the predicted time point by receiving the traffic information of the neighboring roads of the past time and the target road traffic information of the past time point . 11. The method of claim 10,
The collecting of the adjacent roads may include determining whether the adjacent roads are adjacent according to a link depth with the target road, a distance with the target road, a travel time to reach the target road, or a combination thereof The method comprising the steps of: 11. The method of claim 10,
The selecting of the adjacent adjacent roads may be performed,
A method for estimating traffic comprising selecting by support vector machine (SVM) method.

Images